if_lagg.c revision 362368
1/* $OpenBSD: if_trunk.c,v 1.30 2007/01/31 06:20:19 reyk Exp $ */ 2 3/* 4 * Copyright (c) 2005, 2006 Reyk Floeter <reyk@openbsd.org> 5 * Copyright (c) 2007 Andrew Thompson <thompsa@FreeBSD.org> 6 * 7 * Permission to use, copy, modify, and distribute this software for any 8 * purpose with or without fee is hereby granted, provided that the above 9 * copyright notice and this permission notice appear in all copies. 10 * 11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES 12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF 13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR 14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES 15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN 16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF 17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. 18 */ 19 20#include <sys/cdefs.h> 21__FBSDID("$FreeBSD: stable/10/sys/net/if_lagg.c 362368 2020-06-19 07:07:10Z rpokala $"); 22 23#include "opt_inet.h" 24#include "opt_inet6.h" 25 26#include <sys/param.h> 27#include <sys/kernel.h> 28#include <sys/malloc.h> 29#include <sys/mbuf.h> 30#include <sys/queue.h> 31#include <sys/socket.h> 32#include <sys/sockio.h> 33#include <sys/sysctl.h> 34#include <sys/module.h> 35#include <sys/priv.h> 36#include <sys/systm.h> 37#include <sys/proc.h> 38#include <sys/hash.h> 39#include <sys/lock.h> 40#include <sys/rmlock.h> 41#include <sys/taskqueue.h> 42#include <sys/eventhandler.h> 43 44#include <net/ethernet.h> 45#include <net/if.h> 46#include <net/if_clone.h> 47#include <net/if_arp.h> 48#include <net/if_dl.h> 49#include <net/if_llc.h> 50#include <net/if_media.h> 51#include <net/if_types.h> 52#include <net/if_var.h> 53#include <net/bpf.h> 54#include <net/vnet.h> 55 56#if defined(INET) || defined(INET6) 57#include <netinet/in.h> 58#include <netinet/ip.h> 59#endif 60#ifdef INET 61#include <netinet/in_systm.h> 62#include <netinet/if_ether.h> 63#endif 64 65#ifdef INET6 66#include <netinet/ip6.h> 67#include <netinet6/in6_var.h> 68#include <netinet6/in6_ifattach.h> 69#endif 70 71#include <net/if_vlan_var.h> 72#include <net/if_lagg.h> 73#include <net/ieee8023ad_lacp.h> 74 75/* Special flags we should propagate to the lagg ports. */ 76static struct { 77 int flag; 78 int (*func)(struct ifnet *, int); 79} lagg_pflags[] = { 80 {IFF_PROMISC, ifpromisc}, 81 {IFF_ALLMULTI, if_allmulti}, 82 {0, NULL} 83}; 84 85VNET_DEFINE(SLIST_HEAD(__trhead, lagg_softc), lagg_list); /* list of laggs */ 86#define V_lagg_list VNET(lagg_list) 87static VNET_DEFINE(struct mtx, lagg_list_mtx); 88#define V_lagg_list_mtx VNET(lagg_list_mtx) 89#define LAGG_LIST_LOCK_INIT(x) mtx_init(&V_lagg_list_mtx, \ 90 "if_lagg list", NULL, MTX_DEF) 91#define LAGG_LIST_LOCK_DESTROY(x) mtx_destroy(&V_lagg_list_mtx) 92#define LAGG_LIST_LOCK(x) mtx_lock(&V_lagg_list_mtx) 93#define LAGG_LIST_UNLOCK(x) mtx_unlock(&V_lagg_list_mtx) 94eventhandler_tag lagg_detach_cookie = NULL; 95 96static int lagg_clone_create(struct if_clone *, int, caddr_t); 97static void lagg_clone_destroy(struct ifnet *); 98static VNET_DEFINE(struct if_clone *, lagg_cloner); 99#define V_lagg_cloner VNET(lagg_cloner) 100static const char laggname[] = "lagg"; 101 102static void lagg_lladdr(struct lagg_softc *, uint8_t *); 103static void lagg_capabilities(struct lagg_softc *); 104static void lagg_port_lladdr(struct lagg_port *, uint8_t *); 105static void lagg_port_setlladdr(void *, int); 106static int lagg_port_create(struct lagg_softc *, struct ifnet *); 107static int lagg_port_destroy(struct lagg_port *, int); 108static struct mbuf *lagg_input(struct ifnet *, struct mbuf *); 109static void lagg_linkstate(struct lagg_softc *); 110static void lagg_port_state(struct ifnet *, int); 111static int lagg_port_ioctl(struct ifnet *, u_long, caddr_t); 112static int lagg_port_output(struct ifnet *, struct mbuf *, 113 const struct sockaddr *, struct route *); 114static void lagg_port_ifdetach(void *arg __unused, struct ifnet *); 115#ifdef LAGG_PORT_STACKING 116static int lagg_port_checkstacking(struct lagg_softc *); 117#endif 118static void lagg_port2req(struct lagg_port *, struct lagg_reqport *); 119static void lagg_init(void *); 120static void lagg_stop(struct lagg_softc *); 121static int lagg_ioctl(struct ifnet *, u_long, caddr_t); 122static int lagg_ether_setmulti(struct lagg_softc *); 123static int lagg_ether_cmdmulti(struct lagg_port *, int); 124static int lagg_setflag(struct lagg_port *, int, int, 125 int (*func)(struct ifnet *, int)); 126static int lagg_setflags(struct lagg_port *, int status); 127static int lagg_transmit(struct ifnet *, struct mbuf *); 128static void lagg_qflush(struct ifnet *); 129static int lagg_media_change(struct ifnet *); 130static void lagg_media_status(struct ifnet *, struct ifmediareq *); 131static struct lagg_port *lagg_link_active(struct lagg_softc *, 132 struct lagg_port *); 133static const void *lagg_gethdr(struct mbuf *, u_int, u_int, void *); 134 135/* Simple round robin */ 136static void lagg_rr_attach(struct lagg_softc *); 137static int lagg_rr_detach(struct lagg_softc *); 138static int lagg_rr_start(struct lagg_softc *, struct mbuf *); 139static struct mbuf *lagg_rr_input(struct lagg_softc *, struct lagg_port *, 140 struct mbuf *); 141 142/* Active failover */ 143static void lagg_fail_attach(struct lagg_softc *); 144static int lagg_fail_detach(struct lagg_softc *); 145static int lagg_fail_start(struct lagg_softc *, struct mbuf *); 146static struct mbuf *lagg_fail_input(struct lagg_softc *, struct lagg_port *, 147 struct mbuf *); 148 149/* Loadbalancing */ 150static void lagg_lb_attach(struct lagg_softc *); 151static int lagg_lb_detach(struct lagg_softc *); 152static int lagg_lb_port_create(struct lagg_port *); 153static void lagg_lb_port_destroy(struct lagg_port *); 154static int lagg_lb_start(struct lagg_softc *, struct mbuf *); 155static struct mbuf *lagg_lb_input(struct lagg_softc *, struct lagg_port *, 156 struct mbuf *); 157static int lagg_lb_porttable(struct lagg_softc *, struct lagg_port *); 158 159/* 802.3ad LACP */ 160static void lagg_lacp_attach(struct lagg_softc *); 161static int lagg_lacp_detach(struct lagg_softc *); 162static int lagg_lacp_start(struct lagg_softc *, struct mbuf *); 163static struct mbuf *lagg_lacp_input(struct lagg_softc *, struct lagg_port *, 164 struct mbuf *); 165static void lagg_lacp_lladdr(struct lagg_softc *); 166 167static void lagg_callout(void *); 168 169/* lagg protocol table */ 170static const struct lagg_proto { 171 lagg_proto ti_proto; 172 void (*ti_attach)(struct lagg_softc *); 173} lagg_protos[] = { 174 { LAGG_PROTO_ROUNDROBIN, lagg_rr_attach }, 175 { LAGG_PROTO_FAILOVER, lagg_fail_attach }, 176 { LAGG_PROTO_LOADBALANCE, lagg_lb_attach }, 177 { LAGG_PROTO_ETHERCHANNEL, lagg_lb_attach }, 178 { LAGG_PROTO_LACP, lagg_lacp_attach }, 179 { LAGG_PROTO_NONE, NULL } 180}; 181 182SYSCTL_DECL(_net_link); 183SYSCTL_NODE(_net_link, OID_AUTO, lagg, CTLFLAG_RW, 0, 184 "Link Aggregation"); 185 186/* Allow input on any failover links */ 187static VNET_DEFINE(int, lagg_failover_rx_all); 188#define V_lagg_failover_rx_all VNET(lagg_failover_rx_all) 189SYSCTL_INT(_net_link_lagg, OID_AUTO, failover_rx_all, CTLFLAG_RW | CTLFLAG_VNET, 190 &VNET_NAME(lagg_failover_rx_all), 0, 191 "Accept input from any interface in a failover lagg"); 192 193/* Default value for using M_FLOWID */ 194static VNET_DEFINE(int, def_use_flowid) = 1; 195#define V_def_use_flowid VNET(def_use_flowid) 196SYSCTL_INT(_net_link_lagg, OID_AUTO, default_use_flowid, CTLFLAG_RWTUN, 197 &VNET_NAME(def_use_flowid), 0, 198 "Default setting for using flow id for load sharing"); 199 200/* Default value for using M_FLOWID */ 201static VNET_DEFINE(int, def_flowid_shift) = 16; 202#define V_def_flowid_shift VNET(def_flowid_shift) 203SYSCTL_INT(_net_link_lagg, OID_AUTO, default_flowid_shift, CTLFLAG_RWTUN, 204 &VNET_NAME(def_flowid_shift), 0, 205 "Default setting for flowid shift for load sharing"); 206 207static void 208vnet_lagg_init(const void *unused __unused) 209{ 210 211 LAGG_LIST_LOCK_INIT(); 212 SLIST_INIT(&V_lagg_list); 213 V_lagg_cloner = if_clone_simple(laggname, lagg_clone_create, 214 lagg_clone_destroy, 0); 215} 216VNET_SYSINIT(vnet_lagg_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 217 vnet_lagg_init, NULL); 218 219static void 220vnet_lagg_uninit(const void *unused __unused) 221{ 222 223 if_clone_detach(V_lagg_cloner); 224 LAGG_LIST_LOCK_DESTROY(); 225} 226VNET_SYSUNINIT(vnet_lagg_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 227 vnet_lagg_uninit, NULL); 228 229static int 230lagg_modevent(module_t mod, int type, void *data) 231{ 232 233 switch (type) { 234 case MOD_LOAD: 235 lagg_input_p = lagg_input; 236 lagg_linkstate_p = lagg_port_state; 237 lagg_detach_cookie = EVENTHANDLER_REGISTER( 238 ifnet_departure_event, lagg_port_ifdetach, NULL, 239 EVENTHANDLER_PRI_ANY); 240 break; 241 case MOD_UNLOAD: 242 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 243 lagg_detach_cookie); 244 lagg_input_p = NULL; 245 lagg_linkstate_p = NULL; 246 break; 247 default: 248 return (EOPNOTSUPP); 249 } 250 return (0); 251} 252 253static moduledata_t lagg_mod = { 254 "if_lagg", 255 lagg_modevent, 256 0 257}; 258 259DECLARE_MODULE(if_lagg, lagg_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 260MODULE_VERSION(if_lagg, 1); 261 262/* 263 * This routine is run via an vlan 264 * config EVENT 265 */ 266static void 267lagg_register_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 268{ 269 struct lagg_softc *sc = ifp->if_softc; 270 struct lagg_port *lp; 271 struct rm_priotracker tracker; 272 273 if (ifp->if_softc != arg) /* Not our event */ 274 return; 275 276 LAGG_RLOCK(sc, &tracker); 277 if (!SLIST_EMPTY(&sc->sc_ports)) { 278 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 279 EVENTHANDLER_INVOKE(vlan_config, lp->lp_ifp, vtag); 280 } 281 LAGG_RUNLOCK(sc, &tracker); 282} 283 284/* 285 * This routine is run via an vlan 286 * unconfig EVENT 287 */ 288static void 289lagg_unregister_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag) 290{ 291 struct lagg_softc *sc = ifp->if_softc; 292 struct lagg_port *lp; 293 struct rm_priotracker tracker; 294 295 if (ifp->if_softc != arg) /* Not our event */ 296 return; 297 298 LAGG_RLOCK(sc, &tracker); 299 if (!SLIST_EMPTY(&sc->sc_ports)) { 300 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 301 EVENTHANDLER_INVOKE(vlan_unconfig, lp->lp_ifp, vtag); 302 } 303 LAGG_RUNLOCK(sc, &tracker); 304} 305 306static int 307lagg_clone_create(struct if_clone *ifc, int unit, caddr_t params) 308{ 309 struct lagg_softc *sc; 310 struct ifnet *ifp; 311 static const u_char eaddr[6]; /* 00:00:00:00:00:00 */ 312 int i; 313 314 sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_ZERO); 315 ifp = sc->sc_ifp = if_alloc(IFT_ETHER); 316 if (ifp == NULL) { 317 free(sc, M_DEVBUF); 318 return (ENOSPC); 319 } 320 321 sc->sc_ipackets = counter_u64_alloc(M_WAITOK); 322 sc->sc_opackets = counter_u64_alloc(M_WAITOK); 323 sc->sc_ibytes = counter_u64_alloc(M_WAITOK); 324 sc->sc_obytes = counter_u64_alloc(M_WAITOK); 325 326 if (V_def_use_flowid) 327 sc->sc_opts |= LAGG_OPT_USE_FLOWID; 328 sc->flowid_shift = V_def_flowid_shift; 329 330 /* Hash all layers by default */ 331 sc->sc_flags = LAGG_F_HASHL2|LAGG_F_HASHL3|LAGG_F_HASHL4; 332 333 sc->sc_proto = LAGG_PROTO_NONE; 334 for (i = 0; lagg_protos[i].ti_proto != LAGG_PROTO_NONE; i++) { 335 if (lagg_protos[i].ti_proto == LAGG_PROTO_DEFAULT) { 336 sc->sc_proto = lagg_protos[i].ti_proto; 337 lagg_protos[i].ti_attach(sc); 338 break; 339 } 340 } 341 LAGG_LOCK_INIT(sc); 342 LAGG_CALLOUT_LOCK_INIT(sc); 343 SLIST_INIT(&sc->sc_ports); 344 TASK_INIT(&sc->sc_lladdr_task, 0, lagg_port_setlladdr, sc); 345 346 /* 347 * This uses the callout lock rather than the rmlock; one can't 348 * hold said rmlock during SWI. 349 */ 350 callout_init_mtx(&sc->sc_callout, &sc->sc_call_mtx, 0); 351 352 /* Initialise pseudo media types */ 353 ifmedia_init(&sc->sc_media, 0, lagg_media_change, 354 lagg_media_status); 355 ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL); 356 ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO); 357 358 if_initname(ifp, laggname, unit); 359 ifp->if_softc = sc; 360 ifp->if_transmit = lagg_transmit; 361 ifp->if_qflush = lagg_qflush; 362 ifp->if_init = lagg_init; 363 ifp->if_ioctl = lagg_ioctl; 364 ifp->if_flags = IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST; 365 ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS; 366 367 /* 368 * Attach as an ordinary ethernet device, children will be attached 369 * as special device IFT_IEEE8023ADLAG. 370 */ 371 ether_ifattach(ifp, eaddr); 372 373 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 374 lagg_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 375 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 376 lagg_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 377 378 /* Insert into the global list of laggs */ 379 LAGG_LIST_LOCK(); 380 SLIST_INSERT_HEAD(&V_lagg_list, sc, sc_entries); 381 LAGG_LIST_UNLOCK(); 382 383 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 384 385 return (0); 386} 387 388static void 389lagg_clone_destroy(struct ifnet *ifp) 390{ 391 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 392 struct lagg_port *lp; 393 394 LAGG_WLOCK(sc); 395 396 lagg_stop(sc); 397 ifp->if_flags &= ~IFF_UP; 398 399 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach); 400 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach); 401 402 /* Shutdown and remove lagg ports */ 403 while ((lp = SLIST_FIRST(&sc->sc_ports)) != NULL) 404 lagg_port_destroy(lp, 1); 405 /* Unhook the aggregation protocol */ 406 if (sc->sc_detach != NULL) 407 (*sc->sc_detach)(sc); 408 else 409 LAGG_WUNLOCK(sc); 410 411 ifmedia_removeall(&sc->sc_media); 412 ether_ifdetach(ifp); 413 if_free(ifp); 414 415 /* This grabs sc_callout_mtx, serialising it correctly */ 416 callout_drain(&sc->sc_callout); 417 418 /* At this point it's drained; we can free this */ 419 counter_u64_free(sc->sc_ipackets); 420 counter_u64_free(sc->sc_opackets); 421 counter_u64_free(sc->sc_ibytes); 422 counter_u64_free(sc->sc_obytes); 423 424 LAGG_LIST_LOCK(); 425 SLIST_REMOVE(&V_lagg_list, sc, lagg_softc, sc_entries); 426 LAGG_LIST_UNLOCK(); 427 428 taskqueue_drain(taskqueue_swi, &sc->sc_lladdr_task); 429 LAGG_LOCK_DESTROY(sc); 430 LAGG_CALLOUT_LOCK_DESTROY(sc); 431 free(sc, M_DEVBUF); 432} 433 434static void 435lagg_lladdr(struct lagg_softc *sc, uint8_t *lladdr) 436{ 437 struct ifnet *ifp = sc->sc_ifp; 438 struct lagg_port lp; 439 440 if (memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0) 441 return; 442 443 LAGG_WLOCK_ASSERT(sc); 444 /* 445 * Set the link layer address on the lagg interface. 446 * sc_lladdr() notifies the MAC change to 447 * the aggregation protocol. iflladdr_event handler which 448 * may trigger gratuitous ARPs for INET will be handled in 449 * a taskqueue. 450 */ 451 bcopy(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN); 452 if (sc->sc_lladdr != NULL) 453 (*sc->sc_lladdr)(sc); 454 455 bzero(&lp, sizeof(lp)); 456 lp.lp_ifp = sc->sc_ifp; 457 lp.lp_softc = sc; 458 459 lagg_port_lladdr(&lp, lladdr); 460} 461 462static void 463lagg_capabilities(struct lagg_softc *sc) 464{ 465 struct lagg_port *lp; 466 int cap = ~0, ena = ~0; 467 u_long hwa = ~0UL; 468 struct ifnet_hw_tsomax hw_tsomax; 469 470 LAGG_WLOCK_ASSERT(sc); 471 472 memset(&hw_tsomax, 0, sizeof(hw_tsomax)); 473 474 /* Get capabilities from the lagg ports */ 475 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 476 cap &= lp->lp_ifp->if_capabilities; 477 ena &= lp->lp_ifp->if_capenable; 478 hwa &= lp->lp_ifp->if_hwassist; 479 if_hw_tsomax_common(lp->lp_ifp, &hw_tsomax); 480 } 481 cap = (cap == ~0 ? 0 : cap); 482 ena = (ena == ~0 ? 0 : ena); 483 hwa = (hwa == ~0 ? 0 : hwa); 484 485 if (sc->sc_ifp->if_capabilities != cap || 486 sc->sc_ifp->if_capenable != ena || 487 sc->sc_ifp->if_hwassist != hwa || 488 if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax) != 0) { 489 sc->sc_ifp->if_capabilities = cap; 490 sc->sc_ifp->if_capenable = ena; 491 sc->sc_ifp->if_hwassist = hwa; 492 getmicrotime(&sc->sc_ifp->if_lastchange); 493 494 if (sc->sc_ifflags & IFF_DEBUG) 495 if_printf(sc->sc_ifp, 496 "capabilities 0x%08x enabled 0x%08x\n", cap, ena); 497 } 498} 499 500static void 501lagg_port_lladdr(struct lagg_port *lp, uint8_t *lladdr) 502{ 503 struct lagg_softc *sc = lp->lp_softc; 504 struct ifnet *ifp = lp->lp_ifp; 505 struct lagg_llq *llq; 506 int pending = 0; 507 int primary; 508 509 LAGG_WLOCK_ASSERT(sc); 510 511 primary = (sc->sc_primary->lp_ifp == ifp) ? 1 : 0; 512 if (primary == 0 && (lp->lp_detaching || 513 memcmp(lladdr, IF_LLADDR(ifp), ETHER_ADDR_LEN) == 0)) 514 return; 515 516 /* Check to make sure its not already queued to be changed */ 517 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 518 if (llq->llq_ifp == ifp) { 519 pending = 1; 520 break; 521 } 522 } 523 524 if (!pending) { 525 llq = malloc(sizeof(struct lagg_llq), M_DEVBUF, M_NOWAIT); 526 if (llq == NULL) /* XXX what to do */ 527 return; 528 } 529 530 /* Update the lladdr even if pending, it may have changed */ 531 llq->llq_ifp = ifp; 532 llq->llq_primary = primary; 533 bcopy(lladdr, llq->llq_lladdr, ETHER_ADDR_LEN); 534 535 if (!pending) 536 SLIST_INSERT_HEAD(&sc->sc_llq_head, llq, llq_entries); 537 538 taskqueue_enqueue(taskqueue_swi, &sc->sc_lladdr_task); 539} 540 541/* 542 * Set the interface MAC address from a taskqueue to avoid a LOR. 543 */ 544static void 545lagg_port_setlladdr(void *arg, int pending) 546{ 547 struct lagg_softc *sc = (struct lagg_softc *)arg; 548 struct lagg_llq *llq, *head; 549 struct ifnet *ifp; 550 int error; 551 552 /* Grab a local reference of the queue and remove it from the softc */ 553 LAGG_WLOCK(sc); 554 head = SLIST_FIRST(&sc->sc_llq_head); 555 SLIST_FIRST(&sc->sc_llq_head) = NULL; 556 LAGG_WUNLOCK(sc); 557 558 /* 559 * Traverse the queue and set the lladdr on each ifp. It is safe to do 560 * unlocked as we have the only reference to it. 561 */ 562 for (llq = head; llq != NULL; llq = head) { 563 ifp = llq->llq_ifp; 564 565 CURVNET_SET(ifp->if_vnet); 566 if (llq->llq_primary == 0) { 567 /* 568 * Set the link layer address on the laggport interface. 569 * if_setlladdr() triggers gratuitous ARPs for INET. 570 */ 571 error = if_setlladdr(ifp, llq->llq_lladdr, 572 ETHER_ADDR_LEN); 573 if (error) 574 printf("%s: setlladdr failed on %s\n", __func__, 575 ifp->if_xname); 576 } else 577 EVENTHANDLER_INVOKE(iflladdr_event, ifp); 578 CURVNET_RESTORE(); 579 head = SLIST_NEXT(llq, llq_entries); 580 free(llq, M_DEVBUF); 581 } 582} 583 584static int 585lagg_port_create(struct lagg_softc *sc, struct ifnet *ifp) 586{ 587 struct lagg_softc *sc_ptr; 588 struct lagg_port *lp, *tlp; 589 int error = 0; 590 591 LAGG_WLOCK_ASSERT(sc); 592 593 /* Limit the maximal number of lagg ports */ 594 if (sc->sc_count >= LAGG_MAX_PORTS) 595 return (ENOSPC); 596 597 /* Check if port has already been associated to a lagg */ 598 if (ifp->if_lagg != NULL) { 599 /* Port is already in the current lagg? */ 600 lp = (struct lagg_port *)ifp->if_lagg; 601 if (lp->lp_softc == sc) 602 return (EEXIST); 603 return (EBUSY); 604 } 605 606 /* XXX Disallow non-ethernet interfaces (this should be any of 802) */ 607 if (ifp->if_type != IFT_ETHER) 608 return (EPROTONOSUPPORT); 609 610 /* Allow the first Ethernet member to define the MTU */ 611 if (SLIST_EMPTY(&sc->sc_ports)) 612 sc->sc_ifp->if_mtu = ifp->if_mtu; 613 else if (sc->sc_ifp->if_mtu != ifp->if_mtu) { 614 if_printf(sc->sc_ifp, "invalid MTU for %s\n", 615 ifp->if_xname); 616 return (EINVAL); 617 } 618 619 if ((lp = malloc(sizeof(struct lagg_port), 620 M_DEVBUF, M_NOWAIT|M_ZERO)) == NULL) 621 return (ENOMEM); 622 623 /* Check if port is a stacked lagg */ 624 LAGG_LIST_LOCK(); 625 SLIST_FOREACH(sc_ptr, &V_lagg_list, sc_entries) { 626 if (ifp == sc_ptr->sc_ifp) { 627 LAGG_LIST_UNLOCK(); 628 free(lp, M_DEVBUF); 629 return (EINVAL); 630 /* XXX disable stacking for the moment, its untested */ 631#ifdef LAGG_PORT_STACKING 632 lp->lp_flags |= LAGG_PORT_STACK; 633 if (lagg_port_checkstacking(sc_ptr) >= 634 LAGG_MAX_STACKING) { 635 LAGG_LIST_UNLOCK(); 636 free(lp, M_DEVBUF); 637 return (E2BIG); 638 } 639#endif 640 } 641 } 642 LAGG_LIST_UNLOCK(); 643 644 /* Change the interface type */ 645 lp->lp_iftype = ifp->if_type; 646 ifp->if_type = IFT_IEEE8023ADLAG; 647 ifp->if_lagg = lp; 648 lp->lp_ioctl = ifp->if_ioctl; 649 ifp->if_ioctl = lagg_port_ioctl; 650 lp->lp_output = ifp->if_output; 651 ifp->if_output = lagg_port_output; 652 653 lp->lp_ifp = ifp; 654 lp->lp_softc = sc; 655 656 /* Save port link layer address */ 657 bcopy(IF_LLADDR(ifp), lp->lp_lladdr, ETHER_ADDR_LEN); 658 659 if (SLIST_EMPTY(&sc->sc_ports)) { 660 sc->sc_primary = lp; 661 lagg_lladdr(sc, IF_LLADDR(ifp)); 662 } else { 663 /* Update link layer address for this port */ 664 lagg_port_lladdr(lp, IF_LLADDR(sc->sc_ifp)); 665 } 666 667 /* 668 * Insert into the list of ports. 669 * Keep ports sorted by if_index. It is handy, when configuration 670 * is predictable and `ifconfig laggN create ...` command 671 * will lead to the same result each time. 672 */ 673 SLIST_FOREACH(tlp, &sc->sc_ports, lp_entries) { 674 if (tlp->lp_ifp->if_index < ifp->if_index && ( 675 SLIST_NEXT(tlp, lp_entries) == NULL || 676 SLIST_NEXT(tlp, lp_entries)->lp_ifp->if_index > 677 ifp->if_index)) 678 break; 679 } 680 if (tlp != NULL) 681 SLIST_INSERT_AFTER(tlp, lp, lp_entries); 682 else 683 SLIST_INSERT_HEAD(&sc->sc_ports, lp, lp_entries); 684 sc->sc_count++; 685 686 /* Update lagg capabilities */ 687 lagg_capabilities(sc); 688 lagg_linkstate(sc); 689 690 /* Add multicast addresses and interface flags to this port */ 691 lagg_ether_cmdmulti(lp, 1); 692 lagg_setflags(lp, 1); 693 694 if (sc->sc_port_create != NULL) 695 error = (*sc->sc_port_create)(lp); 696 if (error) { 697 /* remove the port again, without calling sc_port_destroy */ 698 lagg_port_destroy(lp, 0); 699 return (error); 700 } 701 702 return (error); 703} 704 705#ifdef LAGG_PORT_STACKING 706static int 707lagg_port_checkstacking(struct lagg_softc *sc) 708{ 709 struct lagg_softc *sc_ptr; 710 struct lagg_port *lp; 711 int m = 0; 712 713 LAGG_WLOCK_ASSERT(sc); 714 715 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 716 if (lp->lp_flags & LAGG_PORT_STACK) { 717 sc_ptr = (struct lagg_softc *)lp->lp_ifp->if_softc; 718 m = MAX(m, lagg_port_checkstacking(sc_ptr)); 719 } 720 } 721 722 return (m + 1); 723} 724#endif 725 726static int 727lagg_port_destroy(struct lagg_port *lp, int runpd) 728{ 729 struct lagg_softc *sc = lp->lp_softc; 730 struct lagg_port *lp_ptr; 731 struct lagg_llq *llq; 732 struct ifnet *ifp = lp->lp_ifp; 733 734 LAGG_WLOCK_ASSERT(sc); 735 736 if (runpd && sc->sc_port_destroy != NULL) 737 (*sc->sc_port_destroy)(lp); 738 739 /* 740 * Remove multicast addresses and interface flags from this port and 741 * reset the MAC address, skip if the interface is being detached. 742 */ 743 if (!lp->lp_detaching) { 744 lagg_ether_cmdmulti(lp, 0); 745 lagg_setflags(lp, 0); 746 lagg_port_lladdr(lp, lp->lp_lladdr); 747 } 748 749 /* Restore interface */ 750 ifp->if_type = lp->lp_iftype; 751 ifp->if_ioctl = lp->lp_ioctl; 752 ifp->if_output = lp->lp_output; 753 ifp->if_lagg = NULL; 754 755 /* Finally, remove the port from the lagg */ 756 SLIST_REMOVE(&sc->sc_ports, lp, lagg_port, lp_entries); 757 sc->sc_count--; 758 759 /* Update the primary interface */ 760 if (lp == sc->sc_primary) { 761 uint8_t lladdr[ETHER_ADDR_LEN]; 762 763 if ((lp_ptr = SLIST_FIRST(&sc->sc_ports)) == NULL) { 764 bzero(&lladdr, ETHER_ADDR_LEN); 765 } else { 766 bcopy(lp_ptr->lp_lladdr, 767 lladdr, ETHER_ADDR_LEN); 768 } 769 lagg_lladdr(sc, lladdr); 770 sc->sc_primary = lp_ptr; 771 772 /* Update link layer address for each port */ 773 SLIST_FOREACH(lp_ptr, &sc->sc_ports, lp_entries) 774 lagg_port_lladdr(lp_ptr, lladdr); 775 } 776 777 /* Remove any pending lladdr changes from the queue */ 778 if (lp->lp_detaching) { 779 SLIST_FOREACH(llq, &sc->sc_llq_head, llq_entries) { 780 if (llq->llq_ifp == ifp) { 781 SLIST_REMOVE(&sc->sc_llq_head, llq, lagg_llq, 782 llq_entries); 783 free(llq, M_DEVBUF); 784 break; /* Only appears once */ 785 } 786 } 787 } 788 789 if (lp->lp_ifflags) 790 if_printf(ifp, "%s: lp_ifflags unclean\n", __func__); 791 792 free(lp, M_DEVBUF); 793 794 /* Update lagg capabilities */ 795 lagg_capabilities(sc); 796 lagg_linkstate(sc); 797 798 return (0); 799} 800 801static int 802lagg_port_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 803{ 804 struct lagg_reqport *rp = (struct lagg_reqport *)data; 805 struct lagg_softc *sc; 806 struct lagg_port *lp = NULL; 807 int error = 0; 808 struct rm_priotracker tracker; 809 810 /* Should be checked by the caller */ 811 if (ifp->if_type != IFT_IEEE8023ADLAG || 812 (lp = ifp->if_lagg) == NULL || (sc = lp->lp_softc) == NULL) 813 goto fallback; 814 815 switch (cmd) { 816 case SIOCGLAGGPORT: 817 if (rp->rp_portname[0] == '\0' || 818 ifunit(rp->rp_portname) != ifp) { 819 error = EINVAL; 820 break; 821 } 822 823 LAGG_RLOCK(sc, &tracker); 824 if ((lp = ifp->if_lagg) == NULL || lp->lp_softc != sc) { 825 error = ENOENT; 826 LAGG_RUNLOCK(sc, &tracker); 827 break; 828 } 829 830 lagg_port2req(lp, rp); 831 LAGG_RUNLOCK(sc, &tracker); 832 break; 833 834 case SIOCSIFCAP: 835 if (lp->lp_ioctl == NULL) { 836 error = EINVAL; 837 break; 838 } 839 error = (*lp->lp_ioctl)(ifp, cmd, data); 840 if (error) 841 break; 842 843 /* Update lagg interface capabilities */ 844 LAGG_WLOCK(sc); 845 lagg_capabilities(sc); 846 LAGG_WUNLOCK(sc); 847 break; 848 849 case SIOCSIFMTU: 850 /* Do not allow the MTU to be changed once joined */ 851 error = EINVAL; 852 break; 853 854 default: 855 goto fallback; 856 } 857 858 return (error); 859 860fallback: 861 if (lp != NULL && lp->lp_ioctl != NULL) 862 return ((*lp->lp_ioctl)(ifp, cmd, data)); 863 864 return (EINVAL); 865} 866 867/* 868 * For direct output to child ports. 869 */ 870static int 871lagg_port_output(struct ifnet *ifp, struct mbuf *m, 872 const struct sockaddr *dst, struct route *ro) 873{ 874 struct lagg_port *lp = ifp->if_lagg; 875 876 switch (dst->sa_family) { 877 case pseudo_AF_HDRCMPLT: 878 case AF_UNSPEC: 879 return ((*lp->lp_output)(ifp, m, dst, ro)); 880 } 881 882 /* drop any other frames */ 883 m_freem(m); 884 return (ENETDOWN); 885} 886 887static void 888lagg_port_ifdetach(void *arg __unused, struct ifnet *ifp) 889{ 890 struct lagg_port *lp; 891 struct lagg_softc *sc; 892 893 if ((lp = ifp->if_lagg) == NULL) 894 return; 895 /* If the ifnet is just being renamed, don't do anything. */ 896 if (ifp->if_flags & IFF_RENAMING) 897 return; 898 899 sc = lp->lp_softc; 900 901 LAGG_WLOCK(sc); 902 lp->lp_detaching = 1; 903 lagg_port_destroy(lp, 1); 904 LAGG_WUNLOCK(sc); 905} 906 907static void 908lagg_port2req(struct lagg_port *lp, struct lagg_reqport *rp) 909{ 910 struct lagg_softc *sc = lp->lp_softc; 911 912 strlcpy(rp->rp_ifname, sc->sc_ifname, sizeof(rp->rp_ifname)); 913 strlcpy(rp->rp_portname, lp->lp_ifp->if_xname, sizeof(rp->rp_portname)); 914 rp->rp_prio = lp->lp_prio; 915 rp->rp_flags = lp->lp_flags; 916 if (sc->sc_portreq != NULL) 917 (*sc->sc_portreq)(lp, (caddr_t)&rp->rp_psc); 918 919 /* Add protocol specific flags */ 920 switch (sc->sc_proto) { 921 case LAGG_PROTO_FAILOVER: 922 if (lp == sc->sc_primary) 923 rp->rp_flags |= LAGG_PORT_MASTER; 924 if (lp == lagg_link_active(sc, sc->sc_primary)) 925 rp->rp_flags |= LAGG_PORT_ACTIVE; 926 break; 927 928 case LAGG_PROTO_ROUNDROBIN: 929 case LAGG_PROTO_LOADBALANCE: 930 case LAGG_PROTO_ETHERCHANNEL: 931 if (LAGG_PORTACTIVE(lp)) 932 rp->rp_flags |= LAGG_PORT_ACTIVE; 933 break; 934 935 case LAGG_PROTO_LACP: 936 /* LACP has a different definition of active */ 937 if (lacp_isactive(lp)) 938 rp->rp_flags |= LAGG_PORT_ACTIVE; 939 if (lacp_iscollecting(lp)) 940 rp->rp_flags |= LAGG_PORT_COLLECTING; 941 if (lacp_isdistributing(lp)) 942 rp->rp_flags |= LAGG_PORT_DISTRIBUTING; 943 break; 944 } 945 946} 947 948static void 949lagg_init(void *xsc) 950{ 951 struct lagg_softc *sc = (struct lagg_softc *)xsc; 952 struct lagg_port *lp; 953 struct ifnet *ifp = sc->sc_ifp; 954 955 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 956 return; 957 958 LAGG_WLOCK(sc); 959 960 ifp->if_drv_flags |= IFF_DRV_RUNNING; 961 /* Update the port lladdrs */ 962 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 963 lagg_port_lladdr(lp, IF_LLADDR(ifp)); 964 965 if (sc->sc_init != NULL) 966 (*sc->sc_init)(sc); 967 968 LAGG_WUNLOCK(sc); 969} 970 971static void 972lagg_stop(struct lagg_softc *sc) 973{ 974 struct ifnet *ifp = sc->sc_ifp; 975 976 LAGG_WLOCK_ASSERT(sc); 977 978 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 979 return; 980 981 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 982 983 if (sc->sc_stop != NULL) 984 (*sc->sc_stop)(sc); 985} 986 987static int 988lagg_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 989{ 990 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 991 struct lagg_reqall *ra = (struct lagg_reqall *)data; 992 struct lagg_reqopts *ro = (struct lagg_reqopts *)data; 993 struct lagg_reqport *rp = (struct lagg_reqport *)data, rpbuf; 994 struct lagg_reqflags *rf = (struct lagg_reqflags *)data; 995 struct ifreq *ifr = (struct ifreq *)data; 996 struct lagg_port *lp; 997 const struct lagg_proto *proto = NULL; 998 struct ifnet *tpif; 999 struct thread *td = curthread; 1000 char *buf, *outbuf; 1001 int count, buflen, len, error = 0; 1002 struct rm_priotracker tracker; 1003 1004 bzero(&rpbuf, sizeof(rpbuf)); 1005 1006 switch (cmd) { 1007 case SIOCGLAGG: 1008 LAGG_RLOCK(sc, &tracker); 1009 count = 0; 1010 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1011 count++; 1012 buflen = count * sizeof(struct lagg_reqport); 1013 LAGG_RUNLOCK(sc, &tracker); 1014 1015 outbuf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 1016 1017 LAGG_RLOCK(sc, &tracker); 1018 ra->ra_proto = sc->sc_proto; 1019 if (sc->sc_req != NULL) 1020 (*sc->sc_req)(sc, (caddr_t)&ra->ra_psc); 1021 1022 count = 0; 1023 buf = outbuf; 1024 len = min(ra->ra_size, buflen); 1025 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1026 if (len < sizeof(rpbuf)) 1027 break; 1028 1029 lagg_port2req(lp, &rpbuf); 1030 memcpy(buf, &rpbuf, sizeof(rpbuf)); 1031 count++; 1032 buf += sizeof(rpbuf); 1033 len -= sizeof(rpbuf); 1034 } 1035 LAGG_RUNLOCK(sc, &tracker); 1036 ra->ra_ports = count; 1037 ra->ra_size = count * sizeof(rpbuf); 1038 error = copyout(outbuf, ra->ra_port, ra->ra_size); 1039 free(outbuf, M_TEMP); 1040 break; 1041 case SIOCSLAGG: 1042 error = priv_check(td, PRIV_NET_LAGG); 1043 if (error) 1044 break; 1045 for (proto = lagg_protos; proto->ti_proto != LAGG_PROTO_NONE; 1046 proto++) { 1047 if (proto->ti_proto == ra->ra_proto) { 1048 if (sc->sc_ifflags & IFF_DEBUG) 1049 printf("%s: using proto %u\n", 1050 sc->sc_ifname, proto->ti_proto); 1051 break; 1052 } 1053 } 1054 if (proto->ti_proto >= LAGG_PROTO_MAX) { 1055 error = EPROTONOSUPPORT; 1056 break; 1057 } 1058 /* Set to LAGG_PROTO_NONE during the attach. */ 1059 LAGG_WLOCK(sc); 1060 if (sc->sc_proto != LAGG_PROTO_NONE) { 1061 int (*sc_detach)(struct lagg_softc *sc); 1062 1063 /* Reset protocol and pointers */ 1064 sc->sc_proto = LAGG_PROTO_NONE; 1065 sc_detach = sc->sc_detach; 1066 sc->sc_detach = NULL; 1067 sc->sc_start = NULL; 1068 sc->sc_input = NULL; 1069 sc->sc_port_create = NULL; 1070 sc->sc_port_destroy = NULL; 1071 sc->sc_linkstate = NULL; 1072 sc->sc_init = NULL; 1073 sc->sc_stop = NULL; 1074 sc->sc_lladdr = NULL; 1075 sc->sc_req = NULL; 1076 sc->sc_portreq = NULL; 1077 1078 if (sc_detach != NULL) 1079 sc_detach(sc); 1080 else 1081 LAGG_WUNLOCK(sc); 1082 } else 1083 LAGG_WUNLOCK(sc); 1084 if (proto->ti_proto != LAGG_PROTO_NONE) 1085 proto->ti_attach(sc); 1086 LAGG_WLOCK(sc); 1087 sc->sc_proto = proto->ti_proto; 1088 LAGG_WUNLOCK(sc); 1089 break; 1090 case SIOCGLAGGOPTS: 1091 ro->ro_opts = sc->sc_opts; 1092 if (sc->sc_proto == LAGG_PROTO_LACP) { 1093 struct lacp_softc *lsc; 1094 1095 lsc = (struct lacp_softc *)sc->sc_psc; 1096 if (lsc->lsc_debug.lsc_tx_test != 0) 1097 ro->ro_opts |= LAGG_OPT_LACP_TXTEST; 1098 if (lsc->lsc_debug.lsc_rx_test != 0) 1099 ro->ro_opts |= LAGG_OPT_LACP_RXTEST; 1100 if (lsc->lsc_strict_mode != 0) 1101 ro->ro_opts |= LAGG_OPT_LACP_STRICT; 1102 if (lsc->lsc_fast_timeout != 0) 1103 ro->ro_opts |= LAGG_OPT_LACP_FAST_TIMO; 1104 1105 ro->ro_active = sc->sc_active; 1106 } else { 1107 ro->ro_active = 0; 1108 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1109 ro->ro_active += LAGG_PORTACTIVE(lp); 1110 } 1111 ro->ro_flapping = sc->sc_flapping; 1112 ro->ro_flowid_shift = sc->flowid_shift; 1113 break; 1114 case SIOCSLAGGOPTS: 1115 error = priv_check(td, PRIV_NET_LAGG); 1116 if (error) 1117 break; 1118 if (ro->ro_opts == 0) 1119 break; 1120 /* 1121 * Set options. LACP options are stored in sc->sc_psc, 1122 * not in sc_opts. 1123 */ 1124 int valid, lacp; 1125 1126 switch (ro->ro_opts) { 1127 case LAGG_OPT_USE_FLOWID: 1128 case -LAGG_OPT_USE_FLOWID: 1129 case LAGG_OPT_FLOWIDSHIFT: 1130 valid = 1; 1131 lacp = 0; 1132 break; 1133 case LAGG_OPT_LACP_TXTEST: 1134 case -LAGG_OPT_LACP_TXTEST: 1135 case LAGG_OPT_LACP_RXTEST: 1136 case -LAGG_OPT_LACP_RXTEST: 1137 case LAGG_OPT_LACP_STRICT: 1138 case -LAGG_OPT_LACP_STRICT: 1139 case LAGG_OPT_LACP_FAST_TIMO: 1140 case -LAGG_OPT_LACP_FAST_TIMO: 1141 valid = lacp = 1; 1142 break; 1143 default: 1144 valid = lacp = 0; 1145 break; 1146 } 1147 1148 LAGG_WLOCK(sc); 1149 if (valid == 0 || 1150 (lacp == 1 && sc->sc_proto != LAGG_PROTO_LACP)) { 1151 /* Invalid combination of options specified. */ 1152 error = EINVAL; 1153 LAGG_WUNLOCK(sc); 1154 break; /* Return from SIOCSLAGGOPTS. */ 1155 } 1156 /* 1157 * Store new options into sc->sc_opts except for 1158 * FLOWIDSHIFT and LACP options. 1159 */ 1160 if (lacp == 0) { 1161 if (ro->ro_opts == LAGG_OPT_FLOWIDSHIFT) 1162 sc->flowid_shift = ro->ro_flowid_shift; 1163 else if (ro->ro_opts > 0) 1164 sc->sc_opts |= ro->ro_opts; 1165 else 1166 sc->sc_opts &= ~ro->ro_opts; 1167 } else { 1168 struct lacp_softc *lsc; 1169 struct lacp_port *lp; 1170 1171 lsc = (struct lacp_softc *)sc->sc_psc; 1172 1173 switch (ro->ro_opts) { 1174 case LAGG_OPT_LACP_TXTEST: 1175 lsc->lsc_debug.lsc_tx_test = 1; 1176 break; 1177 case -LAGG_OPT_LACP_TXTEST: 1178 lsc->lsc_debug.lsc_tx_test = 0; 1179 break; 1180 case LAGG_OPT_LACP_RXTEST: 1181 lsc->lsc_debug.lsc_rx_test = 1; 1182 break; 1183 case -LAGG_OPT_LACP_RXTEST: 1184 lsc->lsc_debug.lsc_rx_test = 0; 1185 break; 1186 case LAGG_OPT_LACP_STRICT: 1187 lsc->lsc_strict_mode = 1; 1188 break; 1189 case -LAGG_OPT_LACP_STRICT: 1190 lsc->lsc_strict_mode = 0; 1191 break; 1192 case LAGG_OPT_LACP_FAST_TIMO: 1193 LACP_LOCK(lsc); 1194 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1195 lp->lp_state |= LACP_STATE_TIMEOUT; 1196 LACP_UNLOCK(lsc); 1197 lsc->lsc_fast_timeout = 1; 1198 break; 1199 case -LAGG_OPT_LACP_FAST_TIMO: 1200 LACP_LOCK(lsc); 1201 LIST_FOREACH(lp, &lsc->lsc_ports, lp_next) 1202 lp->lp_state &= ~LACP_STATE_TIMEOUT; 1203 LACP_UNLOCK(lsc); 1204 lsc->lsc_fast_timeout = 0; 1205 break; 1206 } 1207 } 1208 LAGG_WUNLOCK(sc); 1209 break; 1210 case SIOCGLAGGFLAGS: 1211 rf->rf_flags = sc->sc_flags; 1212 break; 1213 case SIOCSLAGGHASH: 1214 error = priv_check(td, PRIV_NET_LAGG); 1215 if (error) 1216 break; 1217 if ((rf->rf_flags & LAGG_F_HASHMASK) == 0) { 1218 error = EINVAL; 1219 break; 1220 } 1221 LAGG_WLOCK(sc); 1222 sc->sc_flags &= ~LAGG_F_HASHMASK; 1223 sc->sc_flags |= rf->rf_flags & LAGG_F_HASHMASK; 1224 LAGG_WUNLOCK(sc); 1225 break; 1226 case SIOCGLAGGPORT: 1227 if (rp->rp_portname[0] == '\0' || 1228 (tpif = ifunit(rp->rp_portname)) == NULL) { 1229 error = EINVAL; 1230 break; 1231 } 1232 1233 LAGG_RLOCK(sc, &tracker); 1234 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1235 lp->lp_softc != sc) { 1236 error = ENOENT; 1237 LAGG_RUNLOCK(sc, &tracker); 1238 break; 1239 } 1240 1241 lagg_port2req(lp, rp); 1242 LAGG_RUNLOCK(sc, &tracker); 1243 break; 1244 case SIOCSLAGGPORT: 1245 error = priv_check(td, PRIV_NET_LAGG); 1246 if (error) 1247 break; 1248 if (rp->rp_portname[0] == '\0' || 1249 (tpif = ifunit(rp->rp_portname)) == NULL) { 1250 error = EINVAL; 1251 break; 1252 } 1253#ifdef INET6 1254 /* 1255 * A laggport interface should not have inet6 address 1256 * because two interfaces with a valid link-local 1257 * scope zone must not be merged in any form. This 1258 * restriction is needed to prevent violation of 1259 * link-local scope zone. Attempts to add a laggport 1260 * interface which has inet6 addresses triggers 1261 * removal of all inet6 addresses on the member 1262 * interface. 1263 */ 1264 if (in6ifa_llaonifp(tpif)) { 1265 in6_ifdetach(tpif); 1266 if_printf(sc->sc_ifp, 1267 "IPv6 addresses on %s have been removed " 1268 "before adding it as a member to prevent " 1269 "IPv6 address scope violation.\n", 1270 tpif->if_xname); 1271 } 1272#endif 1273 LAGG_WLOCK(sc); 1274 error = lagg_port_create(sc, tpif); 1275 LAGG_WUNLOCK(sc); 1276 break; 1277 case SIOCSLAGGDELPORT: 1278 error = priv_check(td, PRIV_NET_LAGG); 1279 if (error) 1280 break; 1281 if (rp->rp_portname[0] == '\0' || 1282 (tpif = ifunit(rp->rp_portname)) == NULL) { 1283 error = EINVAL; 1284 break; 1285 } 1286 1287 LAGG_WLOCK(sc); 1288 if ((lp = (struct lagg_port *)tpif->if_lagg) == NULL || 1289 lp->lp_softc != sc) { 1290 error = ENOENT; 1291 LAGG_WUNLOCK(sc); 1292 break; 1293 } 1294 1295 error = lagg_port_destroy(lp, 1); 1296 LAGG_WUNLOCK(sc); 1297 break; 1298 case SIOCSIFFLAGS: 1299 /* Set flags on ports too */ 1300 LAGG_WLOCK(sc); 1301 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1302 lagg_setflags(lp, 1); 1303 } 1304 LAGG_WUNLOCK(sc); 1305 1306 if (!(ifp->if_flags & IFF_UP) && 1307 (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1308 /* 1309 * If interface is marked down and it is running, 1310 * then stop and disable it. 1311 */ 1312 LAGG_WLOCK(sc); 1313 lagg_stop(sc); 1314 LAGG_WUNLOCK(sc); 1315 } else if ((ifp->if_flags & IFF_UP) && 1316 !(ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1317 /* 1318 * If interface is marked up and it is stopped, then 1319 * start it. 1320 */ 1321 (*ifp->if_init)(sc); 1322 } 1323 break; 1324 case SIOCADDMULTI: 1325 case SIOCDELMULTI: 1326 LAGG_WLOCK(sc); 1327 error = lagg_ether_setmulti(sc); 1328 LAGG_WUNLOCK(sc); 1329 break; 1330 case SIOCSIFMEDIA: 1331 case SIOCGIFMEDIA: 1332 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1333 break; 1334 1335 case SIOCSIFCAP: 1336 case SIOCSIFMTU: 1337 /* Do not allow the MTU or caps to be directly changed */ 1338 error = EINVAL; 1339 break; 1340 1341 default: 1342 error = ether_ioctl(ifp, cmd, data); 1343 break; 1344 } 1345 return (error); 1346} 1347 1348static int 1349lagg_ether_setmulti(struct lagg_softc *sc) 1350{ 1351 struct lagg_port *lp; 1352 1353 LAGG_WLOCK_ASSERT(sc); 1354 1355 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1356 /* First, remove any existing filter entries. */ 1357 lagg_ether_cmdmulti(lp, 0); 1358 /* copy all addresses from the lagg interface to the port */ 1359 lagg_ether_cmdmulti(lp, 1); 1360 } 1361 return (0); 1362} 1363 1364static int 1365lagg_ether_cmdmulti(struct lagg_port *lp, int set) 1366{ 1367 struct lagg_softc *sc = lp->lp_softc; 1368 struct ifnet *ifp = lp->lp_ifp; 1369 struct ifnet *scifp = sc->sc_ifp; 1370 struct lagg_mc *mc; 1371 struct ifmultiaddr *ifma; 1372 int error; 1373 1374 LAGG_WLOCK_ASSERT(sc); 1375 1376 if (set) { 1377 IF_ADDR_WLOCK(scifp); 1378 TAILQ_FOREACH(ifma, &scifp->if_multiaddrs, ifma_link) { 1379 if (ifma->ifma_addr->sa_family != AF_LINK) 1380 continue; 1381 mc = malloc(sizeof(struct lagg_mc), M_DEVBUF, M_NOWAIT); 1382 if (mc == NULL) { 1383 IF_ADDR_WUNLOCK(scifp); 1384 return (ENOMEM); 1385 } 1386 bcopy(ifma->ifma_addr, &mc->mc_addr, 1387 ifma->ifma_addr->sa_len); 1388 mc->mc_addr.sdl_index = ifp->if_index; 1389 mc->mc_ifma = NULL; 1390 SLIST_INSERT_HEAD(&lp->lp_mc_head, mc, mc_entries); 1391 } 1392 IF_ADDR_WUNLOCK(scifp); 1393 SLIST_FOREACH (mc, &lp->lp_mc_head, mc_entries) { 1394 error = if_addmulti(ifp, 1395 (struct sockaddr *)&mc->mc_addr, &mc->mc_ifma); 1396 if (error) 1397 return (error); 1398 } 1399 } else { 1400 while ((mc = SLIST_FIRST(&lp->lp_mc_head)) != NULL) { 1401 SLIST_REMOVE(&lp->lp_mc_head, mc, lagg_mc, mc_entries); 1402 if (mc->mc_ifma && !lp->lp_detaching) 1403 if_delmulti_ifma(mc->mc_ifma); 1404 free(mc, M_DEVBUF); 1405 } 1406 } 1407 return (0); 1408} 1409 1410/* Handle a ref counted flag that should be set on the lagg port as well */ 1411static int 1412lagg_setflag(struct lagg_port *lp, int flag, int status, 1413 int (*func)(struct ifnet *, int)) 1414{ 1415 struct lagg_softc *sc = lp->lp_softc; 1416 struct ifnet *scifp = sc->sc_ifp; 1417 struct ifnet *ifp = lp->lp_ifp; 1418 int error; 1419 1420 LAGG_WLOCK_ASSERT(sc); 1421 1422 status = status ? (scifp->if_flags & flag) : 0; 1423 /* Now "status" contains the flag value or 0 */ 1424 1425 /* 1426 * See if recorded ports status is different from what 1427 * we want it to be. If it is, flip it. We record ports 1428 * status in lp_ifflags so that we won't clear ports flag 1429 * we haven't set. In fact, we don't clear or set ports 1430 * flags directly, but get or release references to them. 1431 * That's why we can be sure that recorded flags still are 1432 * in accord with actual ports flags. 1433 */ 1434 if (status != (lp->lp_ifflags & flag)) { 1435 error = (*func)(ifp, status); 1436 if (error) 1437 return (error); 1438 lp->lp_ifflags &= ~flag; 1439 lp->lp_ifflags |= status; 1440 } 1441 return (0); 1442} 1443 1444/* 1445 * Handle IFF_* flags that require certain changes on the lagg port 1446 * if "status" is true, update ports flags respective to the lagg 1447 * if "status" is false, forcedly clear the flags set on port. 1448 */ 1449static int 1450lagg_setflags(struct lagg_port *lp, int status) 1451{ 1452 int error, i; 1453 1454 for (i = 0; lagg_pflags[i].flag; i++) { 1455 error = lagg_setflag(lp, lagg_pflags[i].flag, 1456 status, lagg_pflags[i].func); 1457 if (error) 1458 return (error); 1459 } 1460 return (0); 1461} 1462 1463static int 1464lagg_transmit(struct ifnet *ifp, struct mbuf *m) 1465{ 1466 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1467 int error, len, mcast; 1468 struct rm_priotracker tracker; 1469 1470 len = m->m_pkthdr.len; 1471 mcast = (m->m_flags & (M_MCAST | M_BCAST)) ? 1 : 0; 1472 1473 LAGG_RLOCK(sc, &tracker); 1474 /* We need a Tx algorithm and at least one port */ 1475 if (sc->sc_proto == LAGG_PROTO_NONE || sc->sc_count == 0) { 1476 LAGG_RUNLOCK(sc, &tracker); 1477 m_freem(m); 1478 ifp->if_oerrors++; 1479 return (ENXIO); 1480 } 1481 1482 ETHER_BPF_MTAP(ifp, m); 1483 1484 error = (*sc->sc_start)(sc, m); 1485 LAGG_RUNLOCK(sc, &tracker); 1486 1487 if (error == 0) { 1488 counter_u64_add(sc->sc_opackets, 1); 1489 counter_u64_add(sc->sc_obytes, len); 1490 ifp->if_omcasts += mcast; 1491 } else 1492 ifp->if_oerrors++; 1493 1494 return (error); 1495} 1496 1497/* 1498 * The ifp->if_qflush entry point for lagg(4) is no-op. 1499 */ 1500static void 1501lagg_qflush(struct ifnet *ifp __unused) 1502{ 1503} 1504 1505static struct mbuf * 1506lagg_input(struct ifnet *ifp, struct mbuf *m) 1507{ 1508 struct lagg_port *lp = ifp->if_lagg; 1509 struct lagg_softc *sc = lp->lp_softc; 1510 struct ifnet *scifp = sc->sc_ifp; 1511 struct rm_priotracker tracker; 1512 1513 LAGG_RLOCK(sc, &tracker); 1514 if ((scifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 1515 (lp->lp_flags & LAGG_PORT_DISABLED) || 1516 sc->sc_proto == LAGG_PROTO_NONE) { 1517 LAGG_RUNLOCK(sc, &tracker); 1518 m_freem(m); 1519 return (NULL); 1520 } 1521 1522 ETHER_BPF_MTAP(scifp, m); 1523 1524 m = (lp->lp_detaching == 0) ? (*sc->sc_input)(sc, lp, m) : NULL; 1525 1526 if (m != NULL) { 1527 counter_u64_add(sc->sc_ipackets, 1); 1528 counter_u64_add(sc->sc_ibytes, m->m_pkthdr.len); 1529 1530 if (scifp->if_flags & IFF_MONITOR) { 1531 m_freem(m); 1532 m = NULL; 1533 } 1534 } 1535 1536 LAGG_RUNLOCK(sc, &tracker); 1537 return (m); 1538} 1539 1540static int 1541lagg_media_change(struct ifnet *ifp) 1542{ 1543 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1544 1545 if (sc->sc_ifflags & IFF_DEBUG) 1546 printf("%s\n", __func__); 1547 1548 /* Ignore */ 1549 return (0); 1550} 1551 1552static void 1553lagg_media_status(struct ifnet *ifp, struct ifmediareq *imr) 1554{ 1555 struct lagg_softc *sc = (struct lagg_softc *)ifp->if_softc; 1556 struct lagg_port *lp; 1557 struct rm_priotracker tracker; 1558 1559 imr->ifm_status = IFM_AVALID; 1560 imr->ifm_active = IFM_ETHER | IFM_AUTO; 1561 1562 LAGG_RLOCK(sc, &tracker); 1563 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1564 if (LAGG_PORTACTIVE(lp)) 1565 imr->ifm_status |= IFM_ACTIVE; 1566 } 1567 LAGG_RUNLOCK(sc, &tracker); 1568} 1569 1570static void 1571lagg_linkstate(struct lagg_softc *sc) 1572{ 1573 struct lagg_port *lp; 1574 int new_link = LINK_STATE_DOWN; 1575 uint64_t speed; 1576 1577 /* Our link is considered up if at least one of our ports is active */ 1578 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) { 1579 if (lp->lp_link_state == LINK_STATE_UP) { 1580 new_link = LINK_STATE_UP; 1581 break; 1582 } 1583 } 1584 if_link_state_change(sc->sc_ifp, new_link); 1585 1586 /* Update if_baudrate to reflect the max possible speed */ 1587 switch (sc->sc_proto) { 1588 case LAGG_PROTO_FAILOVER: 1589 sc->sc_ifp->if_baudrate = sc->sc_primary != NULL ? 1590 sc->sc_primary->lp_ifp->if_baudrate : 0; 1591 break; 1592 case LAGG_PROTO_ROUNDROBIN: 1593 case LAGG_PROTO_LOADBALANCE: 1594 case LAGG_PROTO_ETHERCHANNEL: 1595 speed = 0; 1596 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1597 speed += lp->lp_ifp->if_baudrate; 1598 sc->sc_ifp->if_baudrate = speed; 1599 break; 1600 case LAGG_PROTO_LACP: 1601 /* LACP updates if_baudrate itself */ 1602 break; 1603 } 1604} 1605 1606static void 1607lagg_port_state(struct ifnet *ifp, int state) 1608{ 1609 struct lagg_port *lp = (struct lagg_port *)ifp->if_lagg; 1610 struct lagg_softc *sc = NULL; 1611 1612 if (lp != NULL) 1613 sc = lp->lp_softc; 1614 if (sc == NULL) 1615 return; 1616 1617 LAGG_WLOCK(sc); 1618 lagg_linkstate(sc); 1619 if (sc->sc_linkstate != NULL) 1620 (*sc->sc_linkstate)(lp); 1621 LAGG_WUNLOCK(sc); 1622} 1623 1624struct lagg_port * 1625lagg_link_active(struct lagg_softc *sc, struct lagg_port *lp) 1626{ 1627 struct lagg_port *lp_next, *rval = NULL; 1628 // int new_link = LINK_STATE_DOWN; 1629 1630 LAGG_RLOCK_ASSERT(sc); 1631 /* 1632 * Search a port which reports an active link state. 1633 */ 1634 1635 if (lp == NULL) 1636 goto search; 1637 if (LAGG_PORTACTIVE(lp)) { 1638 rval = lp; 1639 goto found; 1640 } 1641 if ((lp_next = SLIST_NEXT(lp, lp_entries)) != NULL && 1642 LAGG_PORTACTIVE(lp_next)) { 1643 rval = lp_next; 1644 goto found; 1645 } 1646 1647search: 1648 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1649 if (LAGG_PORTACTIVE(lp_next)) { 1650 rval = lp_next; 1651 goto found; 1652 } 1653 } 1654 1655found: 1656 if (rval != NULL) { 1657 /* 1658 * The IEEE 802.1D standard assumes that a lagg with 1659 * multiple ports is always full duplex. This is valid 1660 * for load sharing laggs and if at least two links 1661 * are active. Unfortunately, checking the latter would 1662 * be too expensive at this point. 1663 XXX 1664 if ((sc->sc_capabilities & IFCAP_LAGG_FULLDUPLEX) && 1665 (sc->sc_count > 1)) 1666 new_link = LINK_STATE_FULL_DUPLEX; 1667 else 1668 new_link = rval->lp_link_state; 1669 */ 1670 } 1671 1672 return (rval); 1673} 1674 1675static const void * 1676lagg_gethdr(struct mbuf *m, u_int off, u_int len, void *buf) 1677{ 1678 if (m->m_pkthdr.len < (off + len)) { 1679 return (NULL); 1680 } else if (m->m_len < (off + len)) { 1681 m_copydata(m, off, len, buf); 1682 return (buf); 1683 } 1684 return (mtod(m, char *) + off); 1685} 1686 1687uint32_t 1688lagg_hashmbuf(struct lagg_softc *sc, struct mbuf *m, uint32_t key) 1689{ 1690 uint16_t etype; 1691 uint32_t p = key; 1692 int off; 1693 struct ether_header *eh; 1694 const struct ether_vlan_header *vlan; 1695#ifdef INET 1696 const struct ip *ip; 1697 const uint32_t *ports; 1698 int iphlen; 1699#endif 1700#ifdef INET6 1701 const struct ip6_hdr *ip6; 1702 uint32_t flow; 1703#endif 1704 union { 1705#ifdef INET 1706 struct ip ip; 1707#endif 1708#ifdef INET6 1709 struct ip6_hdr ip6; 1710#endif 1711 struct ether_vlan_header vlan; 1712 uint32_t port; 1713 } buf; 1714 1715 1716 off = sizeof(*eh); 1717 if (m->m_len < off) 1718 goto out; 1719 eh = mtod(m, struct ether_header *); 1720 etype = ntohs(eh->ether_type); 1721 if (sc->sc_flags & LAGG_F_HASHL2) { 1722 p = hash32_buf(&eh->ether_shost, ETHER_ADDR_LEN, p); 1723 p = hash32_buf(&eh->ether_dhost, ETHER_ADDR_LEN, p); 1724 } 1725 1726 /* Special handling for encapsulating VLAN frames */ 1727 if ((m->m_flags & M_VLANTAG) && (sc->sc_flags & LAGG_F_HASHL2)) { 1728 p = hash32_buf(&m->m_pkthdr.ether_vtag, 1729 sizeof(m->m_pkthdr.ether_vtag), p); 1730 } else if (etype == ETHERTYPE_VLAN) { 1731 vlan = lagg_gethdr(m, off, sizeof(*vlan), &buf); 1732 if (vlan == NULL) 1733 goto out; 1734 1735 if (sc->sc_flags & LAGG_F_HASHL2) 1736 p = hash32_buf(&vlan->evl_tag, sizeof(vlan->evl_tag), p); 1737 etype = ntohs(vlan->evl_proto); 1738 off += sizeof(*vlan) - sizeof(*eh); 1739 } 1740 1741 switch (etype) { 1742#ifdef INET 1743 case ETHERTYPE_IP: 1744 ip = lagg_gethdr(m, off, sizeof(*ip), &buf); 1745 if (ip == NULL) 1746 goto out; 1747 1748 if (sc->sc_flags & LAGG_F_HASHL3) { 1749 p = hash32_buf(&ip->ip_src, sizeof(struct in_addr), p); 1750 p = hash32_buf(&ip->ip_dst, sizeof(struct in_addr), p); 1751 } 1752 if (!(sc->sc_flags & LAGG_F_HASHL4)) 1753 break; 1754 switch (ip->ip_p) { 1755 case IPPROTO_TCP: 1756 case IPPROTO_UDP: 1757 case IPPROTO_SCTP: 1758 iphlen = ip->ip_hl << 2; 1759 if (iphlen < sizeof(*ip)) 1760 break; 1761 off += iphlen; 1762 ports = lagg_gethdr(m, off, sizeof(*ports), &buf); 1763 if (ports == NULL) 1764 break; 1765 p = hash32_buf(ports, sizeof(*ports), p); 1766 break; 1767 } 1768 break; 1769#endif 1770#ifdef INET6 1771 case ETHERTYPE_IPV6: 1772 if (!(sc->sc_flags & LAGG_F_HASHL3)) 1773 break; 1774 ip6 = lagg_gethdr(m, off, sizeof(*ip6), &buf); 1775 if (ip6 == NULL) 1776 goto out; 1777 1778 p = hash32_buf(&ip6->ip6_src, sizeof(struct in6_addr), p); 1779 p = hash32_buf(&ip6->ip6_dst, sizeof(struct in6_addr), p); 1780 flow = ip6->ip6_flow & IPV6_FLOWLABEL_MASK; 1781 p = hash32_buf(&flow, sizeof(flow), p); /* IPv6 flow label */ 1782 break; 1783#endif 1784 } 1785out: 1786 return (p); 1787} 1788 1789int 1790lagg_enqueue(struct ifnet *ifp, struct mbuf *m) 1791{ 1792 1793 return (ifp->if_transmit)(ifp, m); 1794} 1795 1796/* 1797 * Simple round robin aggregation 1798 */ 1799static void 1800lagg_rr_attach(struct lagg_softc *sc) 1801{ 1802 sc->sc_detach = lagg_rr_detach; 1803 sc->sc_start = lagg_rr_start; 1804 sc->sc_input = lagg_rr_input; 1805 sc->sc_detach = NULL; 1806 sc->sc_port_create = NULL; 1807 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1808 sc->sc_seq = 0; 1809} 1810 1811static int 1812lagg_rr_detach(struct lagg_softc *sc) 1813{ 1814 return (0); 1815} 1816 1817static int 1818lagg_rr_start(struct lagg_softc *sc, struct mbuf *m) 1819{ 1820 struct lagg_port *lp; 1821 uint32_t p; 1822 1823 p = atomic_fetchadd_32(&sc->sc_seq, 1); 1824 p %= sc->sc_count; 1825 lp = SLIST_FIRST(&sc->sc_ports); 1826 while (p--) 1827 lp = SLIST_NEXT(lp, lp_entries); 1828 1829 /* 1830 * Check the port's link state. This will return the next active 1831 * port if the link is down or the port is NULL. 1832 */ 1833 if ((lp = lagg_link_active(sc, lp)) == NULL) { 1834 m_freem(m); 1835 return (ENETDOWN); 1836 } 1837 1838 /* Send mbuf */ 1839 return (lagg_enqueue(lp->lp_ifp, m)); 1840} 1841 1842static struct mbuf * 1843lagg_rr_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1844{ 1845 struct ifnet *ifp = sc->sc_ifp; 1846 1847 /* Just pass in the packet to our lagg device */ 1848 m->m_pkthdr.rcvif = ifp; 1849 1850 return (m); 1851} 1852 1853/* 1854 * Active failover 1855 */ 1856static void 1857lagg_fail_attach(struct lagg_softc *sc) 1858{ 1859 sc->sc_detach = lagg_fail_detach; 1860 sc->sc_start = lagg_fail_start; 1861 sc->sc_input = lagg_fail_input; 1862 sc->sc_port_create = NULL; 1863 sc->sc_port_destroy = NULL; 1864 sc->sc_detach = NULL; 1865} 1866 1867static int 1868lagg_fail_detach(struct lagg_softc *sc) 1869{ 1870 return (0); 1871} 1872 1873static int 1874lagg_fail_start(struct lagg_softc *sc, struct mbuf *m) 1875{ 1876 struct lagg_port *lp; 1877 1878 /* Use the master port if active or the next available port */ 1879 if ((lp = lagg_link_active(sc, sc->sc_primary)) == NULL) { 1880 m_freem(m); 1881 return (ENETDOWN); 1882 } 1883 1884 /* Send mbuf */ 1885 return (lagg_enqueue(lp->lp_ifp, m)); 1886} 1887 1888static struct mbuf * 1889lagg_fail_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 1890{ 1891 struct ifnet *ifp = sc->sc_ifp; 1892 struct lagg_port *tmp_tp; 1893 1894 if (lp == sc->sc_primary || V_lagg_failover_rx_all) { 1895 m->m_pkthdr.rcvif = ifp; 1896 return (m); 1897 } 1898 1899 if (!LAGG_PORTACTIVE(sc->sc_primary)) { 1900 tmp_tp = lagg_link_active(sc, sc->sc_primary); 1901 /* 1902 * If tmp_tp is null, we've recieved a packet when all 1903 * our links are down. Weird, but process it anyways. 1904 */ 1905 if ((tmp_tp == NULL || tmp_tp == lp)) { 1906 m->m_pkthdr.rcvif = ifp; 1907 return (m); 1908 } 1909 } 1910 1911 m_freem(m); 1912 return (NULL); 1913} 1914 1915/* 1916 * Loadbalancing 1917 */ 1918static void 1919lagg_lb_attach(struct lagg_softc *sc) 1920{ 1921 struct lagg_port *lp; 1922 struct lagg_lb *lb; 1923 1924 lb = malloc(sizeof(struct lagg_lb), M_DEVBUF, M_WAITOK | M_ZERO); 1925 1926 sc->sc_detach = lagg_lb_detach; 1927 sc->sc_start = lagg_lb_start; 1928 sc->sc_input = lagg_lb_input; 1929 sc->sc_port_create = lagg_lb_port_create; 1930 sc->sc_port_destroy = lagg_lb_port_destroy; 1931 sc->sc_capabilities = IFCAP_LAGG_FULLDUPLEX; 1932 1933 lb->lb_key = arc4random(); 1934 sc->sc_psc = (caddr_t)lb; 1935 1936 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 1937 lagg_lb_port_create(lp); 1938} 1939 1940static int 1941lagg_lb_detach(struct lagg_softc *sc) 1942{ 1943 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1944 LAGG_WUNLOCK(sc); 1945 if (lb != NULL) 1946 free(lb, M_DEVBUF); 1947 return (0); 1948} 1949 1950static int 1951lagg_lb_porttable(struct lagg_softc *sc, struct lagg_port *lp) 1952{ 1953 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1954 struct lagg_port *lp_next; 1955 int i = 0; 1956 1957 bzero(&lb->lb_ports, sizeof(lb->lb_ports)); 1958 SLIST_FOREACH(lp_next, &sc->sc_ports, lp_entries) { 1959 if (lp_next == lp) 1960 continue; 1961 if (i >= LAGG_MAX_PORTS) 1962 return (EINVAL); 1963 if (sc->sc_ifflags & IFF_DEBUG) 1964 printf("%s: port %s at index %d\n", 1965 sc->sc_ifname, lp_next->lp_ifname, i); 1966 lb->lb_ports[i++] = lp_next; 1967 } 1968 1969 return (0); 1970} 1971 1972static int 1973lagg_lb_port_create(struct lagg_port *lp) 1974{ 1975 struct lagg_softc *sc = lp->lp_softc; 1976 return (lagg_lb_porttable(sc, NULL)); 1977} 1978 1979static void 1980lagg_lb_port_destroy(struct lagg_port *lp) 1981{ 1982 struct lagg_softc *sc = lp->lp_softc; 1983 lagg_lb_porttable(sc, lp); 1984} 1985 1986static int 1987lagg_lb_start(struct lagg_softc *sc, struct mbuf *m) 1988{ 1989 struct lagg_lb *lb = (struct lagg_lb *)sc->sc_psc; 1990 struct lagg_port *lp = NULL; 1991 uint32_t p = 0; 1992 1993 if ((sc->sc_opts & LAGG_OPT_USE_FLOWID) && 1994 M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 1995 p = m->m_pkthdr.flowid >> sc->flowid_shift; 1996 else 1997 p = lagg_hashmbuf(sc, m, lb->lb_key); 1998 p %= sc->sc_count; 1999 lp = lb->lb_ports[p]; 2000 2001 /* 2002 * Check the port's link state. This will return the next active 2003 * port if the link is down or the port is NULL. 2004 */ 2005 if ((lp = lagg_link_active(sc, lp)) == NULL) { 2006 m_freem(m); 2007 return (ENETDOWN); 2008 } 2009 2010 /* Send mbuf */ 2011 return (lagg_enqueue(lp->lp_ifp, m)); 2012} 2013 2014static struct mbuf * 2015lagg_lb_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2016{ 2017 struct ifnet *ifp = sc->sc_ifp; 2018 2019 /* Just pass in the packet to our lagg device */ 2020 m->m_pkthdr.rcvif = ifp; 2021 2022 return (m); 2023} 2024 2025/* 2026 * 802.3ad LACP 2027 */ 2028static void 2029lagg_lacp_attach(struct lagg_softc *sc) 2030{ 2031 struct lagg_port *lp; 2032 2033 sc->sc_detach = lagg_lacp_detach; 2034 sc->sc_port_create = lacp_port_create; 2035 sc->sc_port_destroy = lacp_port_destroy; 2036 sc->sc_linkstate = lacp_linkstate; 2037 sc->sc_start = lagg_lacp_start; 2038 sc->sc_input = lagg_lacp_input; 2039 sc->sc_init = lacp_init; 2040 sc->sc_stop = lacp_stop; 2041 sc->sc_lladdr = lagg_lacp_lladdr; 2042 sc->sc_req = lacp_req; 2043 sc->sc_portreq = lacp_portreq; 2044 2045 lacp_attach(sc); 2046 2047 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2048 lacp_port_create(lp); 2049} 2050 2051static int 2052lagg_lacp_detach(struct lagg_softc *sc) 2053{ 2054 struct lagg_port *lp; 2055 void *psc; 2056 2057 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2058 lacp_port_destroy(lp); 2059 2060 psc = sc->sc_psc; 2061 sc->sc_psc = NULL; 2062 LAGG_WUNLOCK(sc); 2063 2064 lacp_detach(psc); 2065 2066 return (0); 2067} 2068 2069static void 2070lagg_lacp_lladdr(struct lagg_softc *sc) 2071{ 2072 struct lagg_port *lp; 2073 2074 /* purge all the lacp ports */ 2075 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2076 lacp_port_destroy(lp); 2077 2078 /* add them back in */ 2079 SLIST_FOREACH(lp, &sc->sc_ports, lp_entries) 2080 lacp_port_create(lp); 2081} 2082 2083static int 2084lagg_lacp_start(struct lagg_softc *sc, struct mbuf *m) 2085{ 2086 struct lagg_port *lp; 2087 2088 lp = lacp_select_tx_port(sc, m); 2089 if (lp == NULL) { 2090 m_freem(m); 2091 return (ENETDOWN); 2092 } 2093 2094 /* Send mbuf */ 2095 return (lagg_enqueue(lp->lp_ifp, m)); 2096} 2097 2098static struct mbuf * 2099lagg_lacp_input(struct lagg_softc *sc, struct lagg_port *lp, struct mbuf *m) 2100{ 2101 struct ifnet *ifp = sc->sc_ifp; 2102 struct ether_header *eh; 2103 u_short etype; 2104 2105 eh = mtod(m, struct ether_header *); 2106 etype = ntohs(eh->ether_type); 2107 2108 /* Tap off LACP control messages */ 2109 if ((m->m_flags & M_VLANTAG) == 0 && etype == ETHERTYPE_SLOW) { 2110 m = lacp_input(lp, m); 2111 if (m == NULL) 2112 return (NULL); 2113 } 2114 2115 /* 2116 * If the port is not collecting or not in the active aggregator then 2117 * free and return. 2118 */ 2119 if (lacp_iscollecting(lp) == 0 || lacp_isactive(lp) == 0) { 2120 m_freem(m); 2121 return (NULL); 2122 } 2123 2124 m->m_pkthdr.rcvif = ifp; 2125 return (m); 2126} 2127 2128static void 2129lagg_callout(void *arg) 2130{ 2131 struct lagg_softc *sc = (struct lagg_softc *)arg; 2132 struct ifnet *ifp = sc->sc_ifp; 2133 2134 ifp->if_ipackets = counter_u64_fetch(sc->sc_ipackets); 2135 ifp->if_opackets = counter_u64_fetch(sc->sc_opackets); 2136 ifp->if_ibytes = counter_u64_fetch(sc->sc_ibytes); 2137 ifp->if_obytes = counter_u64_fetch(sc->sc_obytes); 2138 2139 callout_reset(&sc->sc_callout, hz, lagg_callout, sc); 2140} 2141